/*
[] brackets indicate internal census variable name that can't be disclosed. 

randomly select piks and store the seeds from each run through the loop. 
1) Create an empty base table on which to append the seeds (only do this once before running). 
2) Pull in the seed as a macro variable using the loop counter from the bash script.
3) randomly sample using proc surveyselect, print seed output, and append seed to base table. 
4) Merge back to original data for regressions. 
*/

libname hannah '/projects/users/########/Snapshot2022/IntermediateData';
libname hantmp '/projects/users/########/Snapshot2022/IntermediateData/TempData';
libname ans '/projects/users/########/Snapshot2022/AnalysisData';

/*
* initialize a base empty seed table for appending later seeds;
* ONLY RUN THIS ONCE - DON'T OVERWRITE ON EACH RUN THROUGH BASH LOOP;
proc sql;
  create table hantmp.pikloopseeds
    (
    Label1 char(21) format=$21. informat=$21. label='Label1',
    cValue1 char(13) format=$13. informat=$13. label='cValue1',
    nValue1 num format=D12.3 label='nValue'
    );
quit;
*/

/* 
* get counties merged to individual piks (could do hex IDs);
* ONLY NEED TO DO THIS ONCE;
proc sql;
  create table hantmp.rand_pikscnt as select
  a.pik, b.[county id]
  from hannah.pikspine as a 
    inner join hannah.ehf_icf_spine as b on a.pik=b.pik
  order by a.pik;
quit;
* drop duplicates - get unique piks and their county IDs;
proc sort data = hantmp.rand_pikscnt nodupkey;
  by pik;
run;

* data must be sorted by strata;
proc sort data = hantmp.rand_pikscnt;
  by [county id];
run;
*/

* get frequencies in each strata;
proc freq data = hantmp.rand_pikscnt;
  tables [county id] / out=rand_2freq;
run;

* determine strata sizes. want 1000000 obs. need to define a dataset with that number;
*_nsize_ is the sample size from each strata, positive integer;
* If not a positive integer, deleted and error dataset created;
data rand_2freq2 rand_2error;
  set rand_2freq;
  sampnum=(percent * 1000000)/100;
  _nsize_ = round(sampnum,1);
  sampnum = round(sampnum,0.01);
  if _nsize_ = 0 then output rand_2error;
  if _nsize_ = 0 then delete;
  if count >= ##### then _nsize_ = #####;
  if count < ##### then _nsize_ = count;
output rand_2freq2;
run;
* keep strata variable and sample size in each stratum;
data rand_2freq3;
  set rand_2freq2;
  keep [county id] _nsize_;
run;
data rand_2err3;
  set rand_2error;
  keep [county id] _nsize_;
run;
proc append base = rand_2freq3 data = rand_2err3;
run;
proc sort data = rand_2freq3;
  by [county id];
run;

* Pull in seed from bash loop counter and save as a macro variable;
%let seedi=%sysget(VAR1);

* randomly select one million piks and append the seed to the base table;
proc surveyselect data = hantmp.rand_pikscnt
  out = hantmp.rand_pikloop_lpdid
  sampsize = rand_2freq3
  seed = &seedi;
  strata [county id];
run;

* now need to merge back to full data to make a file Stata can analyze;
proc sql;
  create table ans.rand_pikloop_lpdid as select
  a.*, b.*
  from ans.ehf_icf_analysis as a 
    inner join hantmp.rand_pikloop_lpdid as b on a.pik=b.pik
  order by a.pik, a.year;
quit;


/* 
*For specific groups (e.g., white);
ods output Summary = Summary;
proc surveyselect data = hantmp.white_piks
  out = hantmp.white_pikloop
  method = srs
  sampsize = 1000000
  seed = config outseed;
run;

proc append base = hantmp.pikloopseeds data = summary;
run; quit;
*/

